Building AI tools is different from other software. When you develop an application today, you generally set simple rules, and those rules tend to be static and in code. AI tools, in particular Assistive Intelligence tools, learn from the interactions and from new data continuously.
In addition, if you’ve gone through our training, you would know that Large Language Models are a key building block of these assistive solutions. These building blocks are often produced by other organisations (such as OpenAI) and depending on the setup, sometimes out of your control in terms of their management and retirement. In addition to that, as new models are developed, their usage costs often decrease, and we sometimes see improvements in performance metrics.
This inherent movement in the Large Language Models that are integrated into your AI tools is known as model drift. The name says it all. As time goes by, models need to be updated, continuously checked for accuracy, and just like any other software application, sometimes decommissioned.
